rfreeimage 0.1.0

data/vendor/FreeImage/Source/LibWebP/src/dsp/dsp.yuv_sse2.c

@@ -0,0 +1,322 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// YUV->RGB conversion functions
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./yuv.h"
+
+#if defined(WEBP_USE_SSE2)
+
+#include <emmintrin.h>
+#include <string.h>   // for memcpy
+
+typedef union {   // handy struct for converting SSE2 registers
+  int32_t i32[4];
+  uint8_t u8[16];
+  __m128i m;
+} VP8kCstSSE2;
+
+#if defined(WEBP_YUV_USE_SSE2_TABLES)
+
+#include "./yuv_tables_sse2.h"
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInitSSE2(void) {}
+
+#else
+
+static int done_sse2 = 0;
+static VP8kCstSSE2 VP8kUtoRGBA[256], VP8kVtoRGBA[256], VP8kYtoRGBA[256];
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInitSSE2(void) {
+  if (!done_sse2) {
+    int i;
+    for (i = 0; i < 256; ++i) {
+      VP8kYtoRGBA[i].i32[0] =
+        VP8kYtoRGBA[i].i32[1] =
+        VP8kYtoRGBA[i].i32[2] = (i - 16) * kYScale + YUV_HALF2;
+      VP8kYtoRGBA[i].i32[3] = 0xff << YUV_FIX2;
+
+      VP8kUtoRGBA[i].i32[0] = 0;
+      VP8kUtoRGBA[i].i32[1] = -kUToG * (i - 128);
+      VP8kUtoRGBA[i].i32[2] =  kUToB * (i - 128);
+      VP8kUtoRGBA[i].i32[3] = 0;
+
+      VP8kVtoRGBA[i].i32[0] =  kVToR * (i - 128);
+      VP8kVtoRGBA[i].i32[1] = -kVToG * (i - 128);
+      VP8kVtoRGBA[i].i32[2] = 0;
+      VP8kVtoRGBA[i].i32[3] = 0;
+    }
+    done_sse2 = 1;
+
+#if 0   // code used to generate 'yuv_tables_sse2.h'
+    printf("static const VP8kCstSSE2 VP8kYtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0x%.8x, 0x%.8x, 0x%.8x, 0x%.8x}},\n",
+             VP8kYtoRGBA[i].i32[0], VP8kYtoRGBA[i].i32[1],
+             VP8kYtoRGBA[i].i32[2], VP8kYtoRGBA[i].i32[3]);
+    }
+    printf("};\n\n");
+    printf("static const VP8kCstSSE2 VP8kUtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0, 0x%.8x, 0x%.8x, 0}},\n",
+             VP8kUtoRGBA[i].i32[1], VP8kUtoRGBA[i].i32[2]);
+    }
+    printf("};\n\n");
+    printf("static VP8kCstSSE2 VP8kVtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0x%.8x, 0x%.8x, 0, 0}},\n",
+             VP8kVtoRGBA[i].i32[0], VP8kVtoRGBA[i].i32[1]);
+    }
+    printf("};\n\n");
+#endif
+  }
+}
+
+#endif  // WEBP_YUV_USE_SSE2_TABLES
+
+//-----------------------------------------------------------------------------
+
+static WEBP_INLINE __m128i LoadUVPart(int u, int v) {
+  const __m128i u_part = _mm_loadu_si128(&VP8kUtoRGBA[u].m);
+  const __m128i v_part = _mm_loadu_si128(&VP8kVtoRGBA[v].m);
+  const __m128i uv_part = _mm_add_epi32(u_part, v_part);
+  return uv_part;
+}
+
+static WEBP_INLINE __m128i GetRGBA32bWithUV(int y, const __m128i uv_part) {
+  const __m128i y_part = _mm_loadu_si128(&VP8kYtoRGBA[y].m);
+  const __m128i rgba1 = _mm_add_epi32(y_part, uv_part);
+  const __m128i rgba2 = _mm_srai_epi32(rgba1, YUV_FIX2);
+  return rgba2;
+}
+
+static WEBP_INLINE __m128i GetRGBA32b(int y, int u, int v) {
+  const __m128i uv_part = LoadUVPart(u, v);
+  return GetRGBA32bWithUV(y, uv_part);
+}
+
+static WEBP_INLINE void YuvToRgbSSE2(uint8_t y, uint8_t u, uint8_t v,
+                                     uint8_t* const rgb) {
+  const __m128i tmp0 = GetRGBA32b(y, u, v);
+  const __m128i tmp1 = _mm_packs_epi32(tmp0, tmp0);
+  const __m128i tmp2 = _mm_packus_epi16(tmp1, tmp1);
+  // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp
+  _mm_storel_epi64((__m128i*)rgb, tmp2);
+}
+
+static WEBP_INLINE void YuvToBgrSSE2(uint8_t y, uint8_t u, uint8_t v,
+                                     uint8_t* const bgr) {
+  const __m128i tmp0 = GetRGBA32b(y, u, v);
+  const __m128i tmp1 = _mm_shuffle_epi32(tmp0, _MM_SHUFFLE(3, 0, 1, 2));
+  const __m128i tmp2 = _mm_packs_epi32(tmp1, tmp1);
+  const __m128i tmp3 = _mm_packus_epi16(tmp2, tmp2);
+  // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp
+  _mm_storel_epi64((__m128i*)bgr, tmp3);
+}
+
+//-----------------------------------------------------------------------------
+// Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
+
+#ifdef FANCY_UPSAMPLING
+
+void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst) {
+  int n;
+  for (n = 0; n < 32; n += 4) {
+    const __m128i tmp0_1 = GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]);
+    const __m128i tmp0_2 = GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]);
+    const __m128i tmp0_3 = GetRGBA32b(y[n + 2], u[n + 2], v[n + 2]);
+    const __m128i tmp0_4 = GetRGBA32b(y[n + 3], u[n + 3], v[n + 3]);
+    const __m128i tmp1_1 = _mm_packs_epi32(tmp0_1, tmp0_2);
+    const __m128i tmp1_2 = _mm_packs_epi32(tmp0_3, tmp0_4);
+    const __m128i tmp2 = _mm_packus_epi16(tmp1_1, tmp1_2);
+    _mm_storeu_si128((__m128i*)dst, tmp2);
+    dst += 4 * 4;
+  }
+}
+
+void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst) {
+  int n;
+  for (n = 0; n < 32; n += 2) {
+    const __m128i tmp0_1 = GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]);
+    const __m128i tmp0_2 = GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+  }
+}
+
+void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst) {
+  int n;
+  uint8_t tmp0[2 * 3 + 5 + 15];
+  uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15);  // align
+  for (n = 0; n < 30; ++n) {   // we directly stomp the *dst memory
+    YuvToRgbSSE2(y[n], u[n], v[n], dst + n * 3);
+  }
+  // Last two pixels are special: we write in a tmp buffer before sending
+  // to dst.
+  YuvToRgbSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0);
+  YuvToRgbSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3);
+  memcpy(dst + n * 3, tmp, 2 * 3);
+}
+
+void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst) {
+  int n;
+  uint8_t tmp0[2 * 3 + 5 + 15];
+  uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15);  // align
+  for (n = 0; n < 30; ++n) {
+    YuvToBgrSSE2(y[n], u[n], v[n], dst + n * 3);
+  }
+  YuvToBgrSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0);
+  YuvToBgrSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3);
+  memcpy(dst + n * 3, tmp, 2 * 3);
+}
+
+#endif  // FANCY_UPSAMPLING
+
+//-----------------------------------------------------------------------------
+// Arbitrary-length row conversion functions
+
+static void YuvToRgbaRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 4 <= len; n += 4) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i uv_1 = LoadUVPart(u[1], v[1]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp0_3 = GetRGBA32bWithUV(y[2], uv_1);
+    const __m128i tmp0_4 = GetRGBA32bWithUV(y[3], uv_1);
+    const __m128i tmp1_1 = _mm_packs_epi32(tmp0_1, tmp0_2);
+    const __m128i tmp1_2 = _mm_packs_epi32(tmp0_3, tmp0_4);
+    const __m128i tmp2 = _mm_packus_epi16(tmp1_1, tmp1_2);
+    _mm_storeu_si128((__m128i*)dst, tmp2);
+    dst += 4 * 4;
+    y += 4;
+    u += 2;
+    v += 2;
+  }
+  // Finish off
+  while (n < len) {
+    VP8YuvToRgba(y[0], u[0], v[0], dst);
+    dst += 4;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+    ++n;
+  }
+}
+
+static void YuvToBgraRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 <= len; n += 2) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+    y += 2;
+    ++u;
+    ++v;
+  }
+  // Finish off
+  if (len & 1) {
+    VP8YuvToBgra(y[0], u[0], v[0], dst);
+  }
+}
+
+static void YuvToArgbRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 <= len; n += 2) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(2, 1, 0, 3));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(2, 1, 0, 3));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+    y += 2;
+    ++u;
+    ++v;
+  }
+  // Finish off
+  if (len & 1) {
+    VP8YuvToArgb(y[0], u[0], v[0], dst);
+  }
+}
+
+static void YuvToRgbRowSSE2(const uint8_t* y,
+                            const uint8_t* u, const uint8_t* v,
+                            uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 < len; ++n) {   // we directly stomp the *dst memory
+    YuvToRgbSSE2(y[0], u[0], v[0], dst);  // stomps 8 bytes
+    dst += 3;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+  }
+  VP8YuvToRgb(y[0], u[0], v[0], dst);
+  if (len > 1) {
+    VP8YuvToRgb(y[1], u[n & 1], v[n & 1], dst + 3);
+  }
+}
+
+static void YuvToBgrRowSSE2(const uint8_t* y,
+                            const uint8_t* u, const uint8_t* v,
+                            uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 < len; ++n) {   // we directly stomp the *dst memory
+    YuvToBgrSSE2(y[0], u[0], v[0], dst);  // stomps 8 bytes
+    dst += 3;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+  }
+  VP8YuvToBgr(y[0], u[0], v[0], dst + 0);
+  if (len > 1) {
+    VP8YuvToBgr(y[1], u[n & 1], v[n & 1], dst + 3);
+  }
+}
+
+#endif  // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void WebPInitSamplersSSE2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  WebPSamplers[MODE_RGB]  = YuvToRgbRowSSE2;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRowSSE2;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRowSSE2;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRowSSE2;
+  WebPSamplers[MODE_ARGB] = YuvToArgbRowSSE2;
+#endif  // WEBP_USE_SSE2
+}

data/vendor/FreeImage/Source/LibWebP/src/dsp/lossless.h

@@ -0,0 +1,313 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Image transforms and color space conversion methods for lossless decoder.
+//
+// Authors: Vikas Arora (vikaas.arora@gmail.com)
+//          Jyrki Alakuijala (jyrki@google.com)
+
+#ifndef WEBP_DSP_LOSSLESS_H_
+#define WEBP_DSP_LOSSLESS_H_
+
+#include "../webp/types.h"
+#include "../webp/decode.h"
+
+#include "../enc/histogram.h"
+#include "../utils/utils.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+// Not a trivial literal symbol.
+#define VP8L_NON_TRIVIAL_SYM (0xffffffff)
+
+//------------------------------------------------------------------------------
+// Signatures and generic function-pointers
+
+typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top);
+extern VP8LPredictorFunc VP8LPredictors[16];
+
+typedef void (*VP8LProcessBlueAndRedFunc)(uint32_t* argb_data, int num_pixels);
+extern VP8LProcessBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed;
+extern VP8LProcessBlueAndRedFunc VP8LAddGreenToBlueAndRed;
+
+typedef struct {
+  // Note: the members are uint8_t, so that any negative values are
+  // automatically converted to "mod 256" values.
+  uint8_t green_to_red_;
+  uint8_t green_to_blue_;
+  uint8_t red_to_blue_;
+} VP8LMultipliers;
+typedef void (*VP8LTransformColorFunc)(const VP8LMultipliers* const m,
+                                       uint32_t* argb_data, int num_pixels);
+extern VP8LTransformColorFunc VP8LTransformColor;
+extern VP8LTransformColorFunc VP8LTransformColorInverse;
+
+typedef void (*VP8LConvertFunc)(const uint32_t* src, int num_pixels,
+                                uint8_t* dst);
+extern VP8LConvertFunc VP8LConvertBGRAToRGB;
+extern VP8LConvertFunc VP8LConvertBGRAToRGBA;
+extern VP8LConvertFunc VP8LConvertBGRAToRGBA4444;
+extern VP8LConvertFunc VP8LConvertBGRAToRGB565;
+extern VP8LConvertFunc VP8LConvertBGRAToBGR;
+
+typedef void (*VP8LCollectColorBlueTransformsFunc)(
+    const uint32_t* argb, int stride,
+    int tile_width, int tile_height,
+    int green_to_blue, int red_to_blue, int histo[]);
+extern VP8LCollectColorBlueTransformsFunc VP8LCollectColorBlueTransforms;
+
+typedef void (*VP8LCollectColorRedTransformsFunc)(
+    const uint32_t* argb, int stride,
+    int tile_width, int tile_height,
+    int green_to_red, int histo[]);
+extern VP8LCollectColorRedTransformsFunc VP8LCollectColorRedTransforms;
+
+// Expose some C-only fallback functions
+void VP8LTransformColor_C(const VP8LMultipliers* const m,
+                          uint32_t* data, int num_pixels);
+void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
+                                 uint32_t* data, int num_pixels);
+
+void VP8LConvertBGRAToRGB_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGBA_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGBA4444_C(const uint32_t* src,
+                                 int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGB565_C(const uint32_t* src,
+                               int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToBGR_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LSubtractGreenFromBlueAndRed_C(uint32_t* argb_data, int num_pixels);
+void VP8LAddGreenToBlueAndRed_C(uint32_t* data, int num_pixels);
+
+// Must be called before calling any of the above methods.
+void VP8LDspInit(void);
+
+//------------------------------------------------------------------------------
+// Image transforms.
+
+struct VP8LTransform;  // Defined in dec/vp8li.h.
+
+// Performs inverse transform of data given transform information, start and end
+// rows. Transform will be applied to rows [row_start, row_end[.
+// The *in and *out pointers refer to source and destination data respectively
+// corresponding to the intermediate row (row_start).
+void VP8LInverseTransform(const struct VP8LTransform* const transform,
+                          int row_start, int row_end,
+                          const uint32_t* const in, uint32_t* const out);
+
+// Similar to the static method ColorIndexInverseTransform() that is part of
+// lossless.c, but used only for alpha decoding. It takes uint8_t (rather than
+// uint32_t) arguments for 'src' and 'dst'.
+void VP8LColorIndexInverseTransformAlpha(
+    const struct VP8LTransform* const transform, int y_start, int y_end,
+    const uint8_t* src, uint8_t* dst);
+
+void VP8LResidualImage(int width, int height, int bits, int low_effort,
+                       uint32_t* const argb, uint32_t* const argb_scratch,
+                       uint32_t* const image);
+
+void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
+                             uint32_t* const argb, uint32_t* image);
+
+//------------------------------------------------------------------------------
+// Color space conversion.
+
+// Converts from BGRA to other color spaces.
+void VP8LConvertFromBGRA(const uint32_t* const in_data, int num_pixels,
+                         WEBP_CSP_MODE out_colorspace, uint8_t* const rgba);
+
+//------------------------------------------------------------------------------
+// Misc methods.
+
+// Computes sampled size of 'size' when sampling using 'sampling bits'.
+static WEBP_INLINE uint32_t VP8LSubSampleSize(uint32_t size,
+                                              uint32_t sampling_bits) {
+  return (size + (1 << sampling_bits) - 1) >> sampling_bits;
+}
+
+// -----------------------------------------------------------------------------
+// Faster logarithm for integers. Small values use a look-up table.
+#define LOG_LOOKUP_IDX_MAX 256
+extern const float kLog2Table[LOG_LOOKUP_IDX_MAX];
+extern const float kSLog2Table[LOG_LOOKUP_IDX_MAX];
+typedef float (*VP8LFastLog2SlowFunc)(uint32_t v);
+
+extern VP8LFastLog2SlowFunc VP8LFastLog2Slow;
+extern VP8LFastLog2SlowFunc VP8LFastSLog2Slow;
+
+static WEBP_INLINE float VP8LFastLog2(uint32_t v) {
+  return (v < LOG_LOOKUP_IDX_MAX) ? kLog2Table[v] : VP8LFastLog2Slow(v);
+}
+// Fast calculation of v * log2(v) for integer input.
+static WEBP_INLINE float VP8LFastSLog2(uint32_t v) {
+  return (v < LOG_LOOKUP_IDX_MAX) ? kSLog2Table[v] : VP8LFastSLog2Slow(v);
+}
+
+// -----------------------------------------------------------------------------
+// Huffman-cost related functions.
+
+typedef double (*VP8LCostFunc)(const uint32_t* population, int length);
+typedef double (*VP8LCostCombinedFunc)(const uint32_t* X, const uint32_t* Y,
+                                       int length);
+
+extern VP8LCostFunc VP8LExtraCost;
+extern VP8LCostCombinedFunc VP8LExtraCostCombined;
+
+typedef struct {        // small struct to hold counters
+  int counts[2];        // index: 0=zero steak, 1=non-zero streak
+  int streaks[2][2];    // [zero/non-zero][streak<3 / streak>=3]
+} VP8LStreaks;
+
+typedef VP8LStreaks (*VP8LCostCountFunc)(const uint32_t* population,
+                                         int length);
+typedef VP8LStreaks (*VP8LCostCombinedCountFunc)(const uint32_t* X,
+                                                 const uint32_t* Y, int length);
+
+extern VP8LCostCountFunc VP8LHuffmanCostCount;
+extern VP8LCostCombinedCountFunc VP8LHuffmanCostCombinedCount;
+
+// Get the symbol entropy for the distribution 'population'.
+// Set 'trivial_sym', if there's only one symbol present in the distribution.
+double VP8LPopulationCost(const uint32_t* const population, int length,
+                          uint32_t* const trivial_sym);
+
+// Get the combined symbol entropy for the distributions 'X' and 'Y'.
+double VP8LGetCombinedEntropy(const uint32_t* const X,
+                              const uint32_t* const Y, int length);
+
+// Estimate how many bits the combined entropy of literals and distance
+// approximately maps to.
+double VP8LHistogramEstimateBits(const VP8LHistogram* const p);
+
+// This function estimates the cost in bits excluding the bits needed to
+// represent the entropy code itself.
+double VP8LHistogramEstimateBitsBulk(const VP8LHistogram* const p);
+
+typedef void (*VP8LHistogramAddFunc)(const VP8LHistogram* const a,
+                                     const VP8LHistogram* const b,
+                                     VP8LHistogram* const out);
+extern VP8LHistogramAddFunc VP8LHistogramAdd;
+
+// -----------------------------------------------------------------------------
+// color mapping related functions.
+
+static WEBP_INLINE uint32_t VP8GetARGBIndex(uint32_t idx) {
+  return (idx >> 8) & 0xff;
+}
+
+static WEBP_INLINE uint8_t VP8GetAlphaIndex(uint8_t idx) {
+  return idx;
+}
+
+static WEBP_INLINE uint32_t VP8GetARGBValue(uint32_t val) {
+  return val;
+}
+
+static WEBP_INLINE uint8_t VP8GetAlphaValue(uint32_t val) {
+  return (val >> 8) & 0xff;
+}
+
+typedef void (*VP8LMapARGBFunc)(const uint32_t* src,
+                                const uint32_t* const color_map,
+                                uint32_t* dst, int y_start,
+                                int y_end, int width);
+typedef void (*VP8LMapAlphaFunc)(const uint8_t* src,
+                                 const uint32_t* const color_map,
+                                 uint8_t* dst, int y_start,
+                                 int y_end, int width);
+
+extern VP8LMapARGBFunc VP8LMapColor32b;
+extern VP8LMapAlphaFunc VP8LMapColor8b;
+
+// -----------------------------------------------------------------------------
+// PrefixEncode()
+
+static WEBP_INLINE int VP8LBitsLog2Ceiling(uint32_t n) {
+  const int log_floor = BitsLog2Floor(n);
+  if (n == (n & ~(n - 1)))  // zero or a power of two.
+    return log_floor;
+  else
+    return log_floor + 1;
+}
+
+// Splitting of distance and length codes into prefixes and
+// extra bits. The prefixes are encoded with an entropy code
+// while the extra bits are stored just as normal bits.
+static WEBP_INLINE void VP8LPrefixEncodeBitsNoLUT(int distance, int* const code,
+                                                  int* const extra_bits) {
+  const int highest_bit = BitsLog2Floor(--distance);
+  const int second_highest_bit = (distance >> (highest_bit - 1)) & 1;
+  *extra_bits = highest_bit - 1;
+  *code = 2 * highest_bit + second_highest_bit;
+}
+
+static WEBP_INLINE void VP8LPrefixEncodeNoLUT(int distance, int* const code,
+                                              int* const extra_bits,
+                                              int* const extra_bits_value) {
+  const int highest_bit = BitsLog2Floor(--distance);
+  const int second_highest_bit = (distance >> (highest_bit - 1)) & 1;
+  *extra_bits = highest_bit - 1;
+  *extra_bits_value = distance & ((1 << *extra_bits) - 1);
+  *code = 2 * highest_bit + second_highest_bit;
+}
+
+#define PREFIX_LOOKUP_IDX_MAX   512
+typedef struct {
+  int8_t code_;
+  int8_t extra_bits_;
+} VP8LPrefixCode;
+
+// These tables are derived using VP8LPrefixEncodeNoLUT.
+extern const VP8LPrefixCode kPrefixEncodeCode[PREFIX_LOOKUP_IDX_MAX];
+extern const uint8_t kPrefixEncodeExtraBitsValue[PREFIX_LOOKUP_IDX_MAX];
+static WEBP_INLINE void VP8LPrefixEncodeBits(int distance, int* const code,
+                                             int* const extra_bits) {
+  if (distance < PREFIX_LOOKUP_IDX_MAX) {
+    const VP8LPrefixCode prefix_code = kPrefixEncodeCode[distance];
+    *code = prefix_code.code_;
+    *extra_bits = prefix_code.extra_bits_;
+  } else {
+    VP8LPrefixEncodeBitsNoLUT(distance, code, extra_bits);
+  }
+}
+
+static WEBP_INLINE void VP8LPrefixEncode(int distance, int* const code,
+                                         int* const extra_bits,
+                                         int* const extra_bits_value) {
+  if (distance < PREFIX_LOOKUP_IDX_MAX) {
+    const VP8LPrefixCode prefix_code = kPrefixEncodeCode[distance];
+    *code = prefix_code.code_;
+    *extra_bits = prefix_code.extra_bits_;
+    *extra_bits_value = kPrefixEncodeExtraBitsValue[distance];
+  } else {
+    VP8LPrefixEncodeNoLUT(distance, code, extra_bits, extra_bits_value);
+  }
+}
+
+// In-place difference of each component with mod 256.
+static WEBP_INLINE uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
+  const uint32_t alpha_and_green =
+      0x00ff00ffu + (a & 0xff00ff00u) - (b & 0xff00ff00u);
+  const uint32_t red_and_blue =
+      0xff00ff00u + (a & 0x00ff00ffu) - (b & 0x00ff00ffu);
+  return (alpha_and_green & 0xff00ff00u) | (red_and_blue & 0x00ff00ffu);
+}
+
+void VP8LBundleColorMap(const uint8_t* const row, int width,
+                        int xbits, uint32_t* const dst);
+
+//------------------------------------------------------------------------------
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_DSP_LOSSLESS_H_